Automated Comparison of 3D Images

UCLA researchers in the Department of Radiology have developed an accurate and automatic segmentation algorithm for contouring cerebral blood vessels.

Active contouring is a commonly used image processing method for segmenting cerebral blood vessel aneurysms from computed tomography angiography (CTA) images and evaluating the treatment options. However, conventional active contouring methods such as geodesic active contouring and region completion are limited by: a.) the need to manually input a right set of parameters that may be difficult for users unfamiliar with the algorithm; and b.) inaccurate boundary localization in the regions with poor signal-to-noise ratio (such as bony structures). This is in turn can lead to inaccurate characterization of the aneurysms, and therefore an inaccurate diagnosis.

Dr. Aichi Chen and colleagues at UCLA have developed a novel segmentation algorithm by modifying existing geodesic active contouring method for accurate and automatic contouring of cerebral blood vessels. This algorithm uses an iterative image construction to segment blood vessels, and requires minimal user input. Unlike conventional active contouring methods that segment all the pixels in an image using the same model and GAC parameters, the proposed method segments each individual pixel using separate models and GAC parameters depending on the shape information from only the neighboring pixels. Therefore, the technique is robust to falsely detecting non-vessel components as vessel components.

The image processing method can be used for automatic segmentation of the following pathologies from images acquired using CT and other imaging techniques: